When I moved to ML from neuro I was a little put off by how much less reproducibility was emphasized. Thanks for doing this!
I have been pushing hard for reproducibilty in #ml these are some beginning steps in #icml2019https://twitter.com/rsalakhu/status/1075513718686011404 …
-
-
Thanks. Twitter will use this to make your timeline better. UndoUndo
-
-
-
Not only reproducibility, this also encourages further development! Love this!
Thanks. Twitter will use this to make your timeline better. UndoUndo
-
-
-
IMHO when one publishes/ want to publish a paper they should be asked to submit the code, and/or pretrained model if not the data (although I believe data should also be available) else there is nothing to prove if it at all reproducible And consistent
-
Else there is no difference between a paper and a patent application filed in USPTO, which in my experience requires absolutely no proof of concept as long as anyone else hasn’t claimed it
End of conversation
New conversation -
-
-
Haven't read any recent research papers, but what are your thoughts on also including hardware and training time details in papers? Some papers get published where they trained models across many gpus but that's often not possible for most (?) people when reproducing.
-
Or are you suggesting reproducibility strictly from a verification viewpoint?
End of conversation
New conversation -
Loading seems to be taking a while.
Twitter may be over capacity or experiencing a momentary hiccup. Try again or visit Twitter Status for more information.